DocVLM-Fingerprint: Document VLM Faithfulness Under Visual Perturbations

DocVLM-Fingerprint studies a narrow failure mode in document visual question answering (VQA):a vision-language model (VLM) can return a plausible short answer while its accompanying claimis unsupported by the annotated evidence. This work implements a reproducible evaluation pipelinefor generated chart, optical character recognition (OCR) document, and scientific figure tasks. Thepipeline includes four visual perturbations, open VLM clients, answer scoring, claim-support labels,plots, and failure records. The main run covers 120 generated examples, 480 perturbed variants, andthree open models. Small clean subsets of ChartQA and DocumentVQA, together with aQwen-only CharXiv perturbation supplement using real charts, provide external checks. Theevidence supports a controlled diagnostic study of failure patterns, not a general ranking of VLMs.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-16
DOI
https://doi.org/10.5281/zenodo.24279715
Primary Topic
Multimodal Machine Learning Applications
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

DocVLM-Fingerprint: Document VLM Faithfulness Under Visual Perturbations

Samarth Kukreja
Zenodo (CERN European Organization for Nuclear Research)
Multimodal Machine Learning Applications
preprint

DocVLM-Fingerprint: Document VLM Faithfulness Under Visual Perturbations

Samarth Kukreja
preprint en

Abstract

DocVLM-Fingerprint studies a narrow failure mode in document visual question answering (VQA):a vision-language model (VLM) can return a plausible short answer while its accompanying claimis unsupported by the annotated evidence. This work implements a reproducible evaluation pipelinefor generated chart, optical character recognition (OCR) document, and scientific figure tasks. Thepipeline includes four visual perturbations, open VLM clients, answer scoring, claim-support labels,plots, and failure records. The main run covers 120 generated examples, 480 perturbed variants, andthree open models. Small clean subsets of ChartQA and DocumentVQA, together with aQwen-only CharXiv perturbation supplement using real charts, provide external checks. Theevidence supports a controlled diagnostic study of failure patterns, not a general ranking of VLMs.

Zenodo (CERN European Organization for Nuclear Research)
Multimodal Machine Learning Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon